Zeros of a solution are not usually spaced evenly. A variable coefficient can make an oscillation speed up or slow down, and solving the equation explicitly may be impossible. Nevertheless, the equation can force one solution to have a zero between two zeros of another. Such statements give geometric information without a closed formula.
We first compare two independent solutions of the same equation, then compare solutions of equations with different coefficients. Both arguments use a Wronskian, but the sign of its derivative has a different role in each proof. The results are not the full Sturm–Liouville eigenfunction oscillation theorem; we will state exactly what they establish.
Regular equations and simple zeros
Let be an open interval and suppose , , and are real. Consider
Solutions are real and on . A nonzero solution cannot satisfy : uniqueness for the equivalent first-order system would make it identically zero. Thus every zero is simple. In particular its zeros are isolated: near a zero , .
Definition 2.3.1
(Consecutive zeros)
Two zeros of a solution are consecutive if the solution has no zero in .
Multiplying a nonzero solution by changes neither its zeros nor the equation, and lets us choose its sign on such an interval.
Theorem 2.3.2
(Sturm separation)
If are linearly independent solutions of the same equation and are consecutive zeros of in , then has exactly one zero in . The two solutions have no common zero.
Proof
(The Wronskian gives both existence and uniqueness)
The weighted Wronskian
has derivative zero by the two equations. It is a nonzero constant: if it vanished at one point, the initial-value vectors of and there would be linearly dependent, and uniqueness would make the solutions dependent. In particular, they cannot share a zero.
Choose the sign of so that on . Simplicity gives and . Since
and these are equal and nonzero, and have opposite signs. Continuity gives at least one zero of between them.
For uniqueness, is nonzero in and
has a fixed nonzero sign. Thus is strictly monotone and has at most one zero. This proves separation.
Proof X-ray
Where each hypothesis enters the separation proof
- Constancy of uses only that and solve the same equation. With different coefficients one gets instead, which is why the comparison proof has to track a sign rather than a constant.
- is exactly linear independence, through uniqueness for initial-value vectors. Without it and both steps collapse, as shows.
- Existence of a zero: simplicity gives , and at the endpoints turns the equality into opposite signs of and .
- At most one zero: the identity is valid only where , which is why the interval runs between consecutive zeros of ; positivity of keeps the sign of the derivative fixed on the whole interval.
Comparing different equations
Theorem 2.3.3
(Sturm comparison)
Let the same coefficient be positive in both equations, and let be real with . Let nonzero solutions satisfy
If are consecutive zeros of , then has a zero in . If, in addition, at some point of , then has a zero in the open interval .
Proof
(A sign change in the mixed Wronskian)
For , direct differentiation and the equations give
Choose on , so and . First suppose has no zero in . Change its sign if needed so that there. Then on , but
contradicting nonincrease. Hence there is a zero in the closed interval.
Now suppose at some interior point and has no interior zero. Choose its sign so on ; continuity gives . The endpoint expressions yield . On an open subinterval, continuity gives , and there. Therefore on that subinterval and elsewhere, so , a contradiction. The strict hypothesis upgrades the conclusion to an interior zero.
Proof X-ray
One sign is tracked, and strictness has two sources
The proof is by contradiction and follows only the sign of . Choosing on and fixes the signs of and : simple zeros of give , so would have to rise from to . The hypothesis is used once, to make , so cannot rise. In the closed-interval case, make the endpoint inequalities strict. In the open-interval case they may be equalities, and strictness must come from an open set where . Linear independence is never used.
The comparison is directional: the larger coefficient forces a zero of between the specified zeros of . It does not say that every zero of occurs at a fixed distance from every zero of , nor does it by itself count all eigenfunction zeros.
Consequences and worked examples
Worked example 2.3.4
(Interlacing sine and cosine)
For , take and . Their weighted Wronskian is , so they are independent. The zeros of are , and there is exactly one zero of , namely , in each gap. Conversely, separation also applies with the two functions exchanged.
The theorem does not say every pair of solutions interlaces: and share every zero and are linearly dependent.
Worked example 2.3.5
(An eigenvalue comparison)
Suppose , , and are continuous. If nonzero solve
then their comparison coefficients are . Hence everywhere, and has an interior zero between any consecutive zeros of . For example, on the Dirichlet solutions and correspond to , ; has a zero at between the zeros and of .
Here and extend to the real line, as do both sine solutions; thus the endpoint zeros are interior points of the open comparison interval. This result does not prove that the th eigenfunction has exactly interior zeros; that stronger counting theorem needs additional work.
Worked example 2.3.6
(A bound on zero gaps)
Let be continuous on and . Every nonzero solution of has at least one zero in every closed interval of length : compare it with on . Therefore any two consecutive zeros of are at distance at most : if a gap were longer, a closed interval of that length could be placed strictly inside it.
This bound allows equality, as shown by and . It also guarantees infinitely many zeros in each direction on the real line, because one can choose disjoint intervals of the specified length arbitrarily far away.
Counterexample mode
Why strict comparison changes endpoints
Take , and . The consecutive zeros of at and are the only zeros of in ; there is none in the open interval. Thus the non-strict hypothesis cannot guarantee a zero in the open interval. Equality of coefficients is allowed in the closed-interval conclusion; linear independence is not assumed in the comparison theorem.
The positivity of also matters. For the degenerate equation on , vanishes at zero and is negative on the left half. On either punctured half-interval, and are solutions, but the latter does not extend continuously through zero. The regular uniqueness argument and weighted-Wronskian conclusions cannot simply be carried across that point. The comparison theorems above make no claim for such singular equations.
Common mistakes
Common mistake
(Comparison forces zeros in one direction only)
With , the theorem places a zero of between consecutive zeros of ; it does not place a zero of between consecutive zeros of . For with and with , the consecutive zeros and of enclose no zero of , whose zeros are the multiples of .
Common mistake
(Separation needs consecutive zeros)
The count “exactly one zero” holds between consecutive zeros of . Between zeros that are not consecutive the count can be larger: vanishes at and , and the independent solution has two zeros, and , in between. Apply the theorem gap by gap.
Exercises
Quick check
(For , show that and are independent and identify the zero of between and .)
Solution · Answer
is nonzero. The consecutive zeros of at and contain exactly one zero of , namely .
Quick check
(Suppose on . Bound the gap between consecutive zeros of every nonzero solution of , assuming is continuous. Give an example attaining the bound.)
Solution · Answer
Take in the gap bound: the distance is at most . The solution for attains it.
Quick check
(On an open interval containing , let and let be a nonzero solution of . Prove that has a zero in without solving its equation.)
Solution · Answer
Use and . The inequality is strict at every , so comparison with the consecutive zeros of forces an interior zero of .
Quick check
(Prove that two nonzero real solutions of with a common zero are necessarily linearly dependent.)
Solution · Answer
At the common zero , the initial-value vectors are and . Both derivatives are nonzero because the solutions are nonzero. A suitable scalar multiple has the same initial data as the other solution, and uniqueness makes the functions dependent throughout the interval.